Negative Prompt Examples by Model: SDXL, FLUX & Leonardo
Negative prompt examples for SDXL, FLUX, and Leonardo AI: paste-ready subject kits, FLUX reframes, short Alchemy bans, and mid-2026 honesty.
Turn any photo into an AI prompt — free
No sign-up required. Works with Midjourney, FLUX, DALL-E.
Try Image to Prompt →Negative prompt examples only help when they match the model you paste into. SDXL still wants a short second field. FLUX.2 wants positive substitutes, not a ban wall. Leonardo often exposes a negative box that rewards concise lists. This page is a paste kit by model family: SDXL, FLUX, and Leonardo, with subject variants you can copy today. Soft path when you want SD-leaning positive plus negative blocks from a photo: https://promptmake.net/negative-prompt-generator. You leave with copy-ready strings, conversion rules across hosts, edit habits, and an FAQ. Generator how-tos and long SD listicles ship elsewhere. Stay here for dialect-correct examples you can paste now.
How to use these negative prompt examples
Treat every string below as a starting paste, not a sacred spell. Checkpoint, CFG, sampler, Alchemy settings, and seed change which terms matter. Generate a baseline with an empty or near-empty exclusion first. Add terms for defects you saw. Delete terms that do nothing.
Model routing comes first. If you run Automatic1111 or ComfyUI on SDXL, use the SDXL section. If you run Black Forest Labs FLUX.2 hosts, use the FLUX reframes and ignore classic negative walls. If you run Leonardo with a visible negative field, use the Leonardo section and keep lists short. Mixing dialects burns credits.
PromptMake drafts SD-focused kits from uploads. Soft path: https://promptmake.net/negative-prompt-generator. Convert ideas into Midjourney --no or FLUX prose by hand when those are the paste targets. Guests get about three image generations per day. Registered free accounts get about five.
Who this fits: hybrid creators who switch SDXL, FLUX, and Leonardo in one week and need ready examples per stack. Who should leave: people hunting one mega-list for every generator on earth. That habit failed in 2024 and still fails in mid-2026.
SDXL negative prompt examples (paste-ready)
SDXL in Automatic1111, Forge, and ComfyUI still uses a dedicated negative field. That is the closest thing to classic negative prompt examples that behave as expected. Modern XL bases need less bulk than SD 1.5. Five to fifteen targeted terms beat forty-token quality spam. Weighted negatives like (extra fingers:1.3) help stubborn artifacts after a plain term fails.
Keep CFG in the 5-9 band for most photoreal and illustration work. If a new negative string makes the frame look plastic, drop CFG before you grow the list. Skip EasyNegative and other 1.5 embeddings on XL graphs. Soft path for paired drafts from a reference: https://promptmake.net/negative-prompt-generator.
Paste the positive you already trust. Drop one of the negative strings below into the negative field. Lock seed for the compare. Change one cluster at a time. Save winners with checkpoint name and date so next week's LoRA swap does not inherit stale bans.
Photoreal product and pack shot (SDXL)
Positive skeleton idea: hero product, material callouts, seamless backdrop, key light, soft fill.
Negative paste:
text, watermark, logo, barcode, extra objects, hands, people, frame, border, cartoon, anime, illustration
Add reflection clutter only if messy reflections keep returning. Add (text:1.3) only after plain text failed on a locked seed.
Portrait and headshot (SDXL)
Positive skeleton idea: identity lock, lens cue, key light, clean background or named set.
Negative paste:
extra fingers, fused hands, plastic skin, watermark, text, crowd, second face, cartoon, anime, painting
Skip giant anatomy walls. Change seed before you double the list. For soft film grade in the positive, keep the negative on medium and artifacts so you do not fight grain you asked for.
Interior architecture (SDXL)
Positive skeleton idea: room type, camera height, lens, time of day, materials.
Negative paste:
people, cars, watermark, text, fisheye distortion, oversaturated HDR, cartoon, anime
If furniture clones, fix layout language in the positive before you ban "extra chair" forever.
FLUX negative prompt examples (positive reframes)
Black Forest Labs documents that FLUX.2 does not support negative prompts the way SD does. Official guidance says describe what you want: empty plaza instead of "no crowds," sharp focus instead of "no blur," unmarked surfaces instead of "no text." Some third-party UIs still show a negative box beside FLUX. Treat it as optional and test empty first.
These negative prompt examples for FLUX are therefore positive substitutes. Paste them into the main prompt field as scene language, not into an SD-style negative box. English negation often backfires because the encoder still sees the banned noun. Soft path when you still need an SD dual-field kit for a different host: https://promptmake.net/negative-prompt-generator.
FLUX rewards photographic sentences, lighting detail, and positional clarity. Keep one clear subject. Name materials and light. Prefer empty, solitary, unmarked, and sharp over long without-lists. If defects persist, inpaint or edit with a mask instead of inventing a fake negative wall.
Empty scene and no-crowd reframe (FLUX)
Instead of negative people, crowd, tourists:
Empty plaza at dawn, solitary figure in the middle distance, wide sidewalks clear of pedestrians, quiet city morning, sharp focus throughout
Instead of negative cars, traffic:
Quiet residential street with parked bikes only, empty asphalt, no moving traffic, soft overcast light, documentary photo
Clean product and no-text reframe (FLUX)
Instead of negative text, logo, watermark, label clutter:
Matte black bottle on seamless #F5F5F5 backdrop, unmarked label area, clean glass reflections, softbox key from camera left, product photo, sharp focus
Instead of negative hands, props, clutter:
Single ceramic mug centered on oak table, nothing else in frame, soft window light, shallow depth of field on the handle, still life photo
Portrait clarity reframe (FLUX)
Instead of negative blurry, plastic skin, extra fingers:
Natural skin texture with visible pores, sharp eyes, relaxed hands resting in lap with five clear fingers, soft key light, 85mm portrait photo
Keep FLUX examples in prose. Do not paste SDXL comma walls into Flux 2 Pro, Max, Flex, or Klein hosts and expect Automatic1111 behavior.
For brand color work on FLUX, name the hex in the positive (matte bottle in #1A1A1A) instead of negating rival hues. For typography you want, quote the exact string. For typography you do not want, describe unmarked surfaces and empty margins rather than listing text, letters, alphabet.
Leonardo negative prompt examples
Leonardo often sits between SD-style UIs and style presets. Many modes expose a negative prompt field beside the main prompt. Alchemy and related pipelines reward concise, high-signal bans more than fifty-line walls. Guidance in a moderate band (commonly around 7-9 on classic controls) usually beats cranking guidance as a substitute for a focused list.
Match list length to the model chip you picked that day. SDXL-class fine-tunes on Leonardo tolerate slightly longer descriptive negatives. Alchemy-heavy presets often prefer three to eight precise terms. Photon-class looks can under-generate if you crush the negative too hard. When unsure, start short and grow from measured defects.
PromptMake remains SD-leaning on the dedicated negative path. Soft path: https://promptmake.net/negative-prompt-generator. Use /image when you want Leonardo-targeted positives and then attach one of the short negatives below in Leonardo's own field.
Leonardo photoreal product kit
Negative paste:
blurry, watermark, text, logo, extra objects, hands, cartoon
Optional weight after failure: (text:1.2), (watermark:1.2)
Keep Alchemy on for finals if your plan uses it; explore cheaper without it, then lock the negative that survived exploration.
Leonardo character and portrait kit
Negative paste:
deformed hands, extra fingers, fused fingers, plastic skin, watermark, text, duplicate face
Avoid dumping twenty anatomy synonyms at once. If hands fail, change seed and pose language in the positive before you triple the list.
Leonardo stylized concept kit
Negative paste:
photograph, realistic skin, depth of field, watermark, ui, screenshot, text
When the positive asks for flat illustration or painterly concept art, ban photo cues. When the positive asks for photo, flip the medium cluster the other way.
Convert examples across models without cargo cult
Reuse the idea, not the paste. An SDXL people, cars, watermark kit becomes a FLUX empty-street sentence and a Leonardo three-term negative. A FLUX "unmarked label area" sentence becomes SDXL text, logo, barcode in the negative field. Midjourney users convert the same ideas to --no people, cars, text with three to five terms max.
Never paste FLUX prose into an SDXL negative box and expect CFG magic. Never paste an SDXL wall into FLUX.2 and assume Black Forest Labs wired a second channel. Soft path when the job is SD dual-field from a photo: https://promptmake.net/negative-prompt-generator.
Save three presets per subject: SDXL negative string, FLUX positive reframe, Leonardo short negative. Label the file with model family and date. Teams that keep one shared mega-file burn a week of credits learning the same lesson twice.
When you need a fresh SD dual-field draft from a photo instead of hand-picking an example, generate once and compare against the kits on this page. Soft path: https://promptmake.net/negative-prompt-generator. Keep FLUX and Leonardo conversions in your own preset file so you do not relearn the dialect map every Monday.
Anime and stylized XL fine-tunes on SDXL need a different medium cluster than photoreal bases. Swap photograph, realistic, 3d render into the negative when the positive asks for flat cel or painterly looks. Keep artifact terms. Drop anatomy walls if your fine-tune already draws clean hands on the poses you use.
Common mistakes with negative prompt examples
Mistake: one mega-list for SDXL, FLUX, and Leonardo. Fix: route by model before you paste.
Mistake: treating FLUX UI negative boxes as Automatic1111. Fix: test empty; prefer positive substitutes.
Mistake: expanding Leonardo lists after every failed seed. Fix: change seed, then add one term.
Mistake: negating the main subject. Fix: strengthen the positive.
Mistake: shipping PromptMake output without a skim. Fix: cut unused terms for your checkpoint.
Repair path: empty exclusions, fix the positive, add the smallest ban or reframe that matches a measured failure, regenerate once. Soft path for a fresh SD kit: https://promptmake.net/negative-prompt-generator.
FAQ
Where can I find negative prompt examples for SDXL?
Use the SDXL section above for product, portrait, and interior pastes. Keep five to fifteen terms. Soft path for upload-based kits: https://promptmake.net/negative-prompt-generator. Edit for your checkpoint before a batch.
Do FLUX models need negative prompt examples?
Official FLUX.2 guidance says no classic negatives. Use positive reframes from the FLUX section: empty scenes, unmarked surfaces, sharp focus language. If a host shows a negative box, test it empty before you trust it.
How do Leonardo negative prompt examples differ?
Leonardo often exposes a real negative field but rewards shorter lists than old SD 1.5 walls, especially on Alchemy-style pipelines. Start with three to eight precise terms. Grow only after you measure a defect on that model chip.
Can I reuse the same examples on Midjourney?
Reuse the ideas as a short --no list, not the full SDXL string. Midjourney equates --no to a soft exclusion. Three to five concrete terms beat twenty-token spam. Keep separate presets per family.
How is this different from the generator guide?
The generator guide teaches the upload-to-edit-to-paste loop for PromptMake's SD tool. This page is a catalog of negative prompt examples by model dialect. Use both: generate a draft, then compare against these kits.
How long should negative prompt examples be in 2026?
On SDXL, five to fifteen targeted terms. On Leonardo, often shorter. On FLUX.2, prefer positive sentences over ban lists. Abstract taste words help less than concrete nouns like watermark or frame.
How do I start on PromptMake for free?
Open https://promptmake.net/negative-prompt-generator, upload a reference, and edit the paired blocks for SDXL paste. Guests get about three image runs per day; free accounts about five. Confirm quotas on promptmake.net before a workshop.
Ready to generate your own prompts?
Free. No sign-up required. Works with all major AI models.